When David Lee took on the role of Chief Technology Officer at PwC Ireland, it wasn’t a theoretical exercise. 

Having spent years as a partner advising clients on their digital strategy, he was stepping into the same fire he had helped others walk through.

“Before I took on the role, I was responsible for our day-to-day technology consulting engagements with our clients. I now have that role in terms of setting our own internal direction in the use of technology, including, but not limited to AI,” he says. 

“So in one sense, I now have to eat my own dog food in terms of what I do. And I suppose it’s a very good way to be able to live the experience yourself and put yourself in the shoes of your clients in terms of what they are facing on a day-to-day basis.”

It has also given him a unique perspective on the most hyped technology in decades: artificial intelligence. And if there’s one point he wants to get across, it’s this—Agentic AI is not a panacea.

Lee’s current focus is on understanding—and deploying—this emerging wave of Agentic AI. While GenAI has become familiar in the workplace, he sees Agentic AI as something potentially far more transformational.

“GenAI has been very much focused on the world of intelligent information retrieval and interrogation. Whereas when we start looking at Agentic AI, it takes us into the world of workflow optimisation,” he says. 

“Agentic AI systems are systems that possess the capacity to make autonomous decisions and take actions to achieve specific goals with limited or no direct human intervention.”

This autonomy is what sets Agentic AI apart. “The ability to make decisions in isolation from a human is a feature of agentic-based AI, which would not have been present in what we’ve seen to date from a GenAI perspective,” he says.

Where GenAI boosts individual productivity, Lee says Agentic AI is designed to tackle enterprise-scale challenges. “The goal-oriented behaviour of agentic AI systems provides organisations with the ability to solve more complex problems… rather than doing standalone activities, it allows the linking together of workflows to complete end-to-end processes,” according to Lee.

“It is also multi-agent and system conversation supporting,” he adds. “By that, I mean it has the ability to take data from multiple different sources within our organisation, and parse them together to answer more complex problems.”

Belief and results

Despite the promise, the reality on the ground is more modest. 

PwC’s 2025 CEO survey shows high ambition—but limited results. “Over 90 per cent of Irish CEOs saw their own organisations being changed in terms of their processes and their workflows through AI within the next three-year window,” Lee says. 

“The slight anomaly with that… is that probably only about a fifth of them are able to put their hand on the profitability advantages of using AI to date.”

And while 30 per cent of CEOs believe their companies may not survive the next decade in their current form, AI hasn’t yet delivered a turnaround.

Lee points to five core barriers organisations face in making AI truly work. “If you stand back and try to understand what some of the inhibitors to people making the best advantage of the capabilities are from an AI perspective, they broadly fall under five headings,” he says.

First, he says, is the difficulty of building a business case. “Organisations have made large investments in digitisation of their organisations, with mixed experience in terms of the ability to identify the benefits associated with it. So you’re starting with a business community who are a little bit cautious in terms of the return,” according to Lee.

Second is the skill gap—and it’s not where people often think. “The skills required to take advantage of the technology are quite different from the skills that exist within the organisations currently, and these are not the skills vested in your IT function, but rather the capabilities you need to build into the rest of your workforce,” Lee says.

AI upskilling is also more fluid than other digital transitions. “We’ve all been involved in the introduction of technology where you train people, they use it, and you move on. Whereas the pace of evolution in AI means you train people, they use it, you have to train them again, because the technology has changed.”

Third is trust and governance. “The whole business controls area… if you think of people’s concern over inappropriately using AI in an environment, either with their own employee data or with that of their customers or clients, businesses want to be sure they will not damage what they have,” Lee says. 

Lee emphasises how valuable these long-established trust relationships are. “They’re all immensely proud and aware of the relationships of trust that they’ve developed over the years… So they’re not willing to blindly go forward with the technology, which has yet to be proven in some regards, and trust their business with that,” he says.

Fourth is security. “Unfortunately, Gen AI is not just a tool for the good actors. It’s also available to the bad actors,” Lee warns. “So when we surveyed our CEOs, over 80 per cent of them believe that they will have an increase in cyber risk over the coming 12 months… more than half of them are saying that they will invest more in cyber to protect the organisation from the height of the risk.”

And fifth, Lee points to mindset. “The temptation and the natural place that people gravitate towards is, how do I use this to save cost? If you simply look at it as a tool of efficiency, you’re missing out on some of the more interesting use cases,” he says. 

Cost, revenue and innovation

To Lee, the most exciting use cases for AI aren’t about trimming budgets—they’re about expanding possibility. “There absolutely is a cost advantage to using it. There’s also a revenue gain advantage for using it. No organisation wants to speak to stakeholders about how simply they’re containing costs. It needs to have a narrative in terms of how it increases revenue or enables innovation,” he says.

One healthcare provider PwC works with is applying AI in its radiography function—not to replace radiographers, but to enhance decision-making. “What AI is enabling it to do is to take together image-based data, patient records and lab results, and extrapolate and understand patterns that wouldn’t be immediately obvious to the human,” Lee says.

That has measurable outcomes. “It’s actually reducing the number of unnecessary procedures for patients… and it is reducing the time to result… by 30 per cent,” he says.

Lee also says the new technology has helped the client reimagine how it engages with customers.

“The initial focus, from an AI perspective, was how they could make that process more efficient. How could we support the agents in answering queries? Now you’re beginning to think of people saying, well, actually, it would be far better for everybody if we could use the technology to avoid the complaint in the first instance. So they’re using the technology to try and predict the type of areas which have previously given rise to complaints, so they can get ahead of the game,  have a happier customer, and take out the cost associated with answering queries,” he explains.

That kind of proactive redesign starts with familiarity. “This technology is the one where there’s no substitute for hands-on experience,” he says. “And within our own organisation, we’ve got everyone getting hands-on experience in it, because it begins to get all of our people to understand the capability of the product.” 

Lee shares a recent client story that reflects this mindset shift: “We worked with a client who brought us on board to help them with the adoption of the Microsoft 365 Copilot GenAI solution… getting people to use it in the personal productivity areas.”

But something happened once people got familiar with the platform. “That began to get them thinking about what other applications they could have. And this particular organisation said, actually, we do a lot of work answering quality-related queries from our customers across different geographies,” he says.

“They moved from thinking about it in terms of supporting personal productivity… to building out a very small agent to look over quality documentation, to answer queries that were coming in from customers. I’m not sure they would have arrived at that point if that’s where they started.”

The prize—and the price of getting it wrong

For Lee, Agentic AI is not a quick fix, but a serious platform for transformation, if done right.

“I don’t think that agentic AI is a panacea for some of the challenges that have been faced. I think you must get your upskilling right. You must get your governance right. You must get your controls right,” he says.

“But what agentic AI now gives organisations is the capability to solve more complex problems where the prize is bigger than when I’m simply dealing with the technology that aids individual personal productivity.”

And the greatest danger? “Running before they’ve learned to walk… not giving consideration to cyber… maybe not ambitious enough in terms of the type of problems they try to solve.”

Lee’s message is simple, but pointed: “Stop reading and start doing.”

The Tech Agenda with Ian Kehoe podcast series is sponsored by PwC.